Environmental Monitoring - Part 5 pdf

If it is too small, successive samples will be correlated, inwhich case it will be difficult to determine the confidence interval for the stockestimate.6 The unit of sampling distance ma

Discriminating between the Good and the Bad: Quality Assurance Is Central in Biomonitoring Studies

G Brunialti, P Giordani, and M Ferretti

CONTENTS

20.1 Introduction 444

20.2 Quality Assurance 444

20.3 Errors 445

20.4 Monitoring in a Variable Environment Needs Proper Sampling Design 446

20.4.1 Environmental Factors as Source of Noise in Biomonitoring Data 447

20.4.2 Inherent Variability Requires Unambiguous Objectives 449

20.5 Environmental Factors and Sampling Design 449

20.6 Indicator Development 451

20.7 Observer and Measurement Errors 452

20.7.1 Too Complex or Too Lax Sampling Protocols May Induce Relevant Observer Errors 452

20.7.2 Type of Sampling Measurements 453

20.7.3 Taxonomic Skill of the Operators 454

20.7.4 Observer Error in Lichen Diversity Monitoring 455

20.7.5 Observer Error in Ozone Monitoring with Tobacco Plants 457

20.7.6 Observer Error in Tree Condition Surveys 458

20.7.7 Management of the Learning Factor 459

20.7.8 Time Required for Each Sampling Phase 459

20.8 Conclusions 460

References 460 20

a management is complex, it always depends on the documentation of the varioussteps of the investigations Documentation allows tracking all the steps undertaken

to carry out the investigation of concern and helps in identifying when and whereproblems occur However, documentation can be properly achieved only by ade-quate quality assurance (QA).1,6–8 Politicians, administrators, and decision makersmay be not very interested to know with what degree of confidence a certainpopulation parameter was estimated by the survey they are presenting to the public;for example, colored maps showing the spatial variation of lichen diversity asindicator of air pollution are usually much more attractive than statistical details.However, their attitude may change considerably when the survey results are used,for example, to stop a (supposed) harmful power plant and the owners of the plantchallenge (in terms of money) such a decision in the court In this case, everystatistical detail about the accuracy and precision of the survey data will be verymuch welcome This example suggests that, if biomonitoring should be taken as

a serious basis for decision making, it needs to produce robust, defensible data ofdocumented quality In short, biomonitoring needs QA and it should make thedifferences between “good” (e.g., of documented and therefore known quality)and “bad” (e.g., undocumented and therefore of unknown quality) monitoringprograms

The aim of this chapter is to recall the basic QA procedures, emphasize the needfor a formal design in biomonitoring studies, and provide examples of data qualitycontrol in various fields of environmental biomonitoring, with special reference toair pollution monitoring by means of lichens, sensitive tobacco plants, and sponta-neous vegetation

20.2 QUALITY ASSURANCE

Some definitions of the main activities that have to be carried out in all phases of

a biomonitoring program in order to assess the quality of the data are reported here.Reference will be made to these phases in the rest of the section using the abbrevi-ations given here

QA is an organized group of activities defining the way in which tasks are to

be performed to ensure an expressed level of quality.9 The main benefit of a QAplan is the improved consistency, reliability, and cost-effectiveness of a program

Discriminating between the Good and the Bad 445

through time.8 A QA plan is essential since it forces program managers to identifyand evaluate most of the factors involved in the program In addition, the assessment

of data quality enables mathematical management of uncertainty due to the methodused.8 Cline and Burkman6 consider four main activities in a QA program whichtake all the steps of the monitoring survey into account:

1 Quality Management (QM) This concerns the proper design of the projectand its documentation It ensures that the proper activities are performed

in the proper way QM activities include, for example, the choice of theproper sampling strategy to be adopted (i.e., where and how the samplingstations have to be located)

2 Quality Assurance (QA) This concerns the first steps of evaluation ofthe quality of the data It includes the use and documentation ofstandard operating procedures All the activities defined in the samplingprotocol are examples of QA procedures In the case of lichen moni-toring, these activities include the selection of sampling subplots, theselection of standard trees, and the positioning of the sampling grid onthe tree

3 Quality Control (QC) This concerns mostly the training, calibration, andcontrol phases It ensures that data are collected appropriately and that

QA is carried out

4 Quality Evaluation (QE) This concerns mainly the statistical evaluation

of the data quality These activities enable the precision and accuracy ofthe data collected by the operators to be evaluated, providing the basisfor comparability of the data

20.3 ERRORS

Environmental data for large areas are generally assessed by sample-based methods.The objective of a sample-based survey is to select a subset — the sample — fromthe population of interest and to estimate population parameters based on probabilitytheory.10 Obviously, these parameter estimates differ from the true population asthey are subject to different sources of errors.5

Errors can be classified into four major categories:2,5,11

1 Sampling errors: These errors are generated by the nature of the samplingitself and by the degree of data variability In general, sampling errors can

be reduced by increasing the sample size and by introducing a more efficient sampling design.2,5,11

cost-2 Assessment errors: These errors incorporate measurement and tion errors They can occur when the methodology is poorly standardized,when insufficient care is devoted to its application, or when there areproblems with instrument calibration.2

classifica-3 Prediction errors: Many attributes in environmental resources assessmentare not directly assessed but derived by models In this case it is assumedthat based on the input values the true population value is derived Models

In order to improve the interpretation of survey results and to review the benefit

of the retrieved information, the total error of estimates has to be quantified Someauthors introduced the terms “error budget”5,12,13 or “total sampling design”14 todefine this parameter The error budget provides a calculation of the total erroraffecting the survey estimates, which can be achieved by a mathematical model thataccounts for the various error sources A general parameter of the studied populationadopted to calculate the total survey error is the mean square error of an estimate.Köhl et al.5 report the following formula by Kish12:

(20.1)

where is the mean square error, ∑r S r2 is the sum of all variance terms (S r)from multiple error sources, and (∑r B r) the squared sum of the biases (B r)

20.4 MONITORING IN A VARIABLE ENVIRONMENT

NEEDS PROPER SAMPLING DESIGN

According to Yoccoz et al.,15 we can define monitoring as the process of gatheringinformation about some system-state variables at different points in time for the purpose

of assessing the state of the system and making inferences about changes in stateover time If our focus is on the monitoring of biological diversity, the systems ofinterest to us are typically ecosystems or components of such systems (e.g., com-munities and populations), and the variables of interest include quantities such asspecies richness, species diversity, biomass, and population size.15

In the assessment of environmental quality by means of biomonitors, it is important

to control the variability of biological data, which often affects the forecastingprecision of these techniques.16 According to Kovacs,3 the quality of the data orig-inating from biological measurements depends heavily on at least three factors: (1)variability of the biomonitoring organisms (interactions between the organisms andenvironmental factors), (2) operators involved in data collection (especially for methodsrequiring taxonomic knowledge), and (3) type of sampling (sampling design, density

of sampling points)

The selection of a proper (suitable) sampling design represents the first step toreduce data variability due to sampling error When selecting the proper design, theobjectives of the survey and environmental variability should be taken carefully intoaccount

2

MSE(y)

Discriminating between the Good and the Bad 447

20.4.1 E NVIRONMENTAL F ACTORS AS S OURCE OF N OISE

IN B IOMONITORING D ATA

Environmental factors such as geomorphology, climatic variables, and substratecould have a great impact on the ecosystem property being studied in order to assessenvironmental quality, such as the rate of indicator species, the biodiversity of acommunity, or the presence of injuries on organisms For this reason, it is important

to understand the environmental processes and the interactions underlying thechanges they undergo

Large-scale monitoring programs cover large geographical regions and raise thequestion of how to deal with the great differences among ecosystems found in thevarious areas.17

Different species show various patterns of geographical distribution irrespective

of anthropogenic effects on the environment; neither is it reasonable to expect theirranges of distribution to be constant over time even in the absence of human activities.The factors that determine the ranges of distribution and geographical patterns inspecies diversity are not well understood, and, in some cases, it is therefore difficult

to separate the natural pattern of variations from the effects of human activities.17Due to this variability, study of environmental properties with known cause–effect relationship is to be preferred Indeed, the higher the potential to isolate thecause, the lesser is the error in interpreting the data For some methodologies thecause–effect relationship is more obvious, for example, in the case of ecotoxicolog-ical experiments or in the case of ozone-sensitive tobacco, in which noise is easilyidentified since the cultivar Bel W3 is preferentially sensitive to ozone-related inju-ries For other types of biomonitoring, it is more difficult to discriminate betweenthe influence of a single variable and that of the others

With reference, above all, to biodiversity studies, we should remember thatenvironmental processes are dynamic: populations of organisms are in a constantstate of flux Intensive small-scale studies to calibrate the response of subsets of allspecies will enhance understanding of the generality and predictability of the trendsindicated by the response of subsets.18 This type of approach enables a valid modelfor interpreting data obtained on a large scale to be developed.18 Several examples

19 carried out a study aimed at standardizing LichenBiodiversity monitoring in a Mediterranean region In particular, they analyzed theinfluence of the great geomorphological variability and of substrate characteristics

on epiphytic lichen vegetation The results obtained in a small area suggested theuse of less restrictive parameters in the sampling protocol (namely, olive trees with

an inclination of the trunk >30˚ were also suitable for biomonitoring relevés) Furtherinformation on the applicability of this method was obtained by means of an in-depth study on three tree species monitored in the same microclimatic conditions,showing that the results obtained in the sample area could be extended to vasterareas.20 In general, these investigations are useful in the preliminary stages ofpreparation of the sampling protocol and for developing models enabling the infor-mation obtained to be extended on a large scale A wide array of models has beendeveloped to cover aspects as diverse as biogeography, conservation biology, climate

of such an approach are present in the literature (see, for example, References 19,

20, and 21) Giordani et al

448 Environmental Monitoring

change research, and habitat or species management Conceptual considerationsshould relate to selecting appropriate procedure for model selection Testing themodel in a wider range of situations (in space and time) will enable the range ofapplications for which the model predictions are suitable to be defined.22

An important issue is selectivity which seems particularly important in ecologicalmeasurement A protocol is selective if the response provided as a measurementdepends only on the intended ecosystem property.17 Regarding this aspect, Yoccoz

et al.15 suggest that quantitative state variables characterizing the system well should

be privileged For example, when defining management objectives in terms of changes

of densities of indicator species, the program should incorporate tests to ensure thatselected species are indeed indicators of the process and variables of interest.15From the point of view of application, some examples of processing for limitingerrors due to environmental variability in interpreting ILB values (Index of LichenBiodiversity) are reported by Ferretti and Erhardt2 and by Loppi et al.23 According

to this method, the ILB values are calculated as percentage deviations from ral/normal conditions (Table 20.1), i.e., from a maximum ILB potentially measurable

natu-in a natural area In Italy, ILB values are natu-interpreted accordnatu-ing to different scales21,23depending on the bioclimatic regions,24 determined on the basis of the distribution

of indicator lichen species and of the main meteorological and climatic parameters(rainfall, altitude, and temperature) This approach was also used in other types ofbiomonitoring, as, for example, in the case of bioaccumulation of trace elements inlichens.25–27 The advantage of this interpretation is that it enables data measured on

TABLE 20.1

Interpretative Scales for Index of Lichen Biodiversity Values Scored

in the Humid Sub-Mediterranean Bioclimatic Region (Thyrrenian Italy) and in the Dry Sub-Mediterranean Bioclimatic Region (Adriatic)

in Italy

% Deviation from

Natural Condition

IBL Score (Thyrrenian Italy)

IBL Score (Adriatic Italy)

Naturality/Alteration Classes

Note: The scales are based on percentage deviation from maximum score potentially assessed

in background natural conditions.

Source: Modified from Loppi, S et al., A new scale for the interpretation of lichen biodiversity values in the Thyrrenian side of Italy, in Progress and Problems in Lichenology at the Turn

of the Millennium, 4th IAL Symposium (IAL 4), Llimona, X., Lumbsh, H.T., and Ott, S., Eds.,

Bibliotheca Lichenologica, 82, J Cramer in der Gebruder Borntraeger Verlagsbuchhandlung, Berlin, Stuttgart, 2002, 235 and unpublished data 21

Discriminating between the Good and the Bad 449

a regional scale to be compared with data on a national scale Inevitably, however,these developments are only approximations that do not always lead to a reduction

in error, since they tend to simplify the effects of the interactions between organismsand the environment For this reason, it is important that the quality level of the datathat can actually be achieved should be consistent with the level of predictabilitysuggested by the interpretation of the results Some authors,28,29 for example, haveobserved different bioaccumulation rates in different lichen species collected underthe same environmental conditions As a consequence, the combined use of differentspecies in the same bioaccumulation survey has to be verified previously to checkthe correlation among elemental concentration in the accumulator species Examples

of calibration of interpretation scales in different bioclimatic regions are also reported

in other field of environmental monitoring such as in bioaccumulation in mosses26

or in macroinvertebrates for the assessment of fresh water quality.30

20.4.2 I NHERENT V ARIABILITY R EQUIRES U NAMBIGUOUS O BJECTIVES

An explicit and well-defined objective is the major driver of the whole designprocess.2 Once the nature of the study is identified, the unambiguous definition ofthe objectives involves the explicit identification of:

• Assessment question: Careful attention should be paid in the phase ofdefinition of the objectives of the program Different objectives requiredifferent monitoring designs.31 As a consequence, the scope of inference

of the study and the data collected depend on the aim of the study If themonitoring objectives are clearly stated, it will be easier to describe thestatistical methods to be used to analyze the data.17

• Target population: Unfortunately, monitoring programs do not alwaysdefine their target population in an explicit statement In many cases, thestatement is insufficiently clear to determine whether a potential sampleunit is included or excluded from the target population.17

• Geographical coverage: The area to be considered by the investigation,

as also the characteristic of the area, are important when considering theproper sampling design; e.g., large vs small survey areas or flat vs.geomorphological complex areas

20.5 ENVIRONMENTAL FACTORS AND SAMPLING

DESIGN

As previously reported, depending on the objectives of the investigation, mental data for large geographical units are generally collected by sample surveys.The objective is therefore to select a subset from the population of interest (thesample) that allows inferences about the whole population

environ-A good sampling design is essential to collect data amenable to statisticalanalyses and to control errors in relation to the costs

Many environmental programs address the assessment of abundance and richness

as state variables of interest.15 However, when these results have to be extended to

Similar results were obtained by Humphrey et al.33 in a study to assess thedifferences in the species richness of lichen and bryophyte communities betweenplanted and seminatural stands A high percentage of species was recorded onlyonce and very few species were common to more than half the plots This “localrarity” phenomenon has been noted in other studies34–36 and is partially related tosampling area The authors of that study observe that it is possible that a 1-hasampling plot used is too small to capture a representative sample of lower plantdiversity in forest stands For example, Rose37 recommends a minimum samplingarea of 1 km2, but again, this depends on the objective of the survey.

Recently, the influence of different sampling tactics in the evaluation of lichenbiodiversity was performed in a test study in Italy In the first sampling method(Table 20.2), the five trees nearest to the center of the sampling unit were selected,within a Primary Sampling Unit of 1 km2 In the second sample, the operator movedfrom the center of the square in all of the cardinal directions, took a circular plotwith radius of 56 m (Secondary Sampling Units — SSUs) and selected (if there were)

TABLE 20.2

Sampling Procedures in Four Methods for Assessing Lichen Biodiversity

Sampling Tactic Plot Dimension and Shape

No of Trees and Selection Procedures

Method 1 1 km 2 primary unit The 5 trees nearest to the center

of the primary unit Method 2 4 circular subplots (secondary

sampling units — SSU) with 56-m radius, within 1 km 2 primary unit

The 3 trees nearest to the center

of each plot

Method 3 4 circular subplots (secondary

sampling units — SSU) with 125-m radius, within 1 km 2 primary unit

The 3 trees nearest to the center

of each plot

Method 4 10 circular random plots with

30-m radius, within 1 km 2 primary unit

All the trees within the plots

Note: Details in the text.

Discriminating between the Good and the Bad 451

the three trees nearest to the center of each plot In the third sampling tactic, the

SSUs had a radius of 125 m Finally, in the fourth method, used to obtain the true

average ILB value, a random selection of 30-m-radius SSU was adopted In each

secondary unit a census of the trees within the plot was carried out

Significant differences in average ILB values were found between the checked

tactics As a result, the third tactic gives the better estimation of the average ILB of

the sampling units and it can find a sufficient number of sampling trees, whereas a

56-m-radius SSU is too small to find a proper number of trees The first tactic (five

trees nearest the center) often takes to a “clustering error,” i.e., the trees selected are

grouped in a small portion of the sampling units and are not representative of the

whole area

Another important aspect to consider is the sampling density, which needs to

be defined in relation to the objectives of the study and to its spatial scale Ferretti

et al.38 used two datasets of lichen diversity (LD) surveys at the subnational level

in Italy for establishing the sampling density that can be cost-effectively adopted in

medium- to large-scale biomonitoring studies As expected, in both cases the relative

error on the mean LD values and the error associated with the interpolation of LD

values for (unmeasured) grid cells increase with decreasing sampling density

How-ever, it was possible to identify sampling density able to provide acceptable errors

with quite a strong reduction of sampling efforts This is important, as reduction of

sampling effort can result in a considerable saving of resources that can be used for

more detailed investigation in potentially problematic areas

20.6 INDICATOR DEVELOPMENT

We can define “indicator” as a character or an entity that can be measured to estimate

status and trends of the target environmental resource.39 Further, an “index” is a

characteristic, usually expressed as a score, that describes the status of an indicator.8

Response indicators should demonstrate the following features:2,39–41

• Correlate with changes in processes or other unmeasured components

such as the stressor of concern

• Have a broad application, both at local and at large scale

• Integrate effects over time

• Provide early warning on future changes in ecosystem condition

• Provide distinctive information, e.g., cause–effect

• Be related to the overall structure and function of ecosystems

• Have a low and standard measurement error

• Have a sufficient reference on the effective applicability on the field

• Be cost effective

The development of indicator and indices is important in environmental

moni-toring39 above all to obtain concise information from a complex environment The

process for indicator development should be taken into account including all the

phases of the sampling, from the assessment question to the selection of core

indicators and the evaluation of the performance of the indicators adopted

452 Environmental Monitoring

For this reason it is important to establish a priori the variables of interest in a

sampling protocol

An example of a rigorous selection of the variables to be measured is given by

EMAN, the Ecological Monitoring and Assessment Network implemented in

Can-ada.42 In this program a core of suitable variables capable of identifying departures

from normal ranges of fluctuations in key ecosystem parameters were selected to

detect early warning of ecosystem change The criteria for this selection were based

on data quality, applicability, data collection, repeatability, data analysis and

inter-pretation, and cost-effectiveness

20.7 OBSERVER AND MEASUREMENT ERRORS

In implementing the guidelines for biomonitoring methods, the documentation of

standard operating procedures and a proper sampling design are only the first steps

towards meeting Quality Assurance Objectives.6

To assess the reliability and consistency of the data, two activities above all are

fundamental: training of the personnel involved in data collection and field checks

on reproducibility of data.43 Observer and measurement errors are important issues,

especially in large-scale and long-term studies involving many surveyors and

sub-jective estimates of a given attribute.43–45 Metric measurements, often considered in

forest health assessment (dbh, distance between trees, etc.), depend closely on the

precision of the instrument used and they are generally easily repeatable and

repro-ducible When considering methods based on visual estimation, the instrument is

the human eye Visual assessments are quickly made and do not require expensive

equipment, chemical tests, or highly trained personnel, but their subjective nature

is a matter for concern.46 Measurements based on visual estimation are consequently

less precise and repeatable and less accurately reproducible

Many different factors may contribute to the variability of application of a visually

based assessment method: the operational manual used and the accuracy and precision

of the operator, which in turn define his/her position on the calibration curve

The main causes of error due to the operators involved in biomonitoring surveys

imental protocol used: the wrong type of measurement and imprecision of the

instruments, the need for more experience in taxonomy, and faulty timing of the

various phases

20.7.1 T OO C OMPLEX OR T OO L AX S AMPLING P ROTOCOLS MAY

INDUCE RELEVANT OBSERVER ERRORS

A sampling protocol that calls for excessively complex procedures might not be

easily applicable on a large scale and by inexperienced operators The

applica-bility of a protocol can be assessed by measuring precision and reproduciapplica-bility

in the framework of resampling surveys The development of preparatory studies,

conducted perhaps on a small scale and in controlled conditions, is a good method

for assessing both the applicability and the repeatability of an experimental protocol.47

are reported in Table 20.3 It is possible to distinguish errors relating to the

exper-An excessively lax sampling protocol leaving too much to the operator’s jectivity, would, on the other hand, contribute towards increasing the sampling errorand make it difficult to control data quality Again, in this case, the error induced

sub-by the operator’s subjectivity can be assessed sub-by repeating every phase of the sampling

in order to identify which phase is most influenced by the error

20.7.2 TYPE OF SAMPLING MEASUREMENTS

Efficient field sampling protocols strive to achieve high metric precision becausethis ensures the repeatability of observations among crews.47

Simplifying, we can distinguish between direct and indirect measurements such

as visual estimates This second type of measurement may be desirable in somecases because of its rapidity, but low precision or loss of ecological information maylimit its application.47

As already mentioned, the use of quantitative state variables is recommended

in order to reduce the error in data collecting due to the subjectivity of the operators.However, this is not always possible in environmental studies because of the greatvariety of parameters involved

The evaluation of metric precision for a study of the riparian forest surveys gested that objectively assessed metrics such as tree/snag DBH have high precision,

sug-TABLE 20.3

Main Observer Errors and Procedures to Improve Data Quality

Factors of Errors Due to

Operators Evaluation Criteria

Procedures to Improve Data Quality

Too complex sampling

protocol (scarce applicability)

Precision Reproducibility

Propedeutic studies Field checks Too rough sampling protocol

(high level of subjectivity)

Accuracy Precision Reproducibility

Propedeutic studies Field checks

Kind of measures (e.g.,

nonquantitative measures,

inadequate instruments)

Precision Reproducibility

Rigorous protocol Application in the field

Taxonomic knowledge Intercalibration tests

Accuracy Bias estimation Audit certification

Training Certification Debriefing Harmonization

Harmonization Time for each phase of the

sampling protocol

Precision Reproducibility

Rigorous protocol Time restrictions

Note: See text for details.

while estimates such as woody debris or tree-cover metrics are less precise Barker47showed that the tree DBH measurements were very precise because they were meas-ured with a diameter tape, while crown ratio measures were less precise because theywere visually assessed.48

The advantage of visual assessments is that they are quick to evaluate and donot require expensive equipment, chemical tests, or highly trained personnel, butthe trade-off is frequently loss of ecological detail.47 Another example is provided

by Englund et al.49 A comparison among different canopy-density measurementtechniques showed that hemispherical photography is a more versatile techniquethan the spherical densiometer, but the equipment is more expensive, analysis time

is not trivial, and waiting for appropriate conditions for taking pictures is a significantlimitation Finally, we must consider that the ultimate use of the data should dictatewhether the increased precision that would result from more exacting measures is

an acceptable trade-off for the investment in time and required degree of ecologicalinformation.47

20.7.3 TAXONOMIC SKILL OF THE OPERATORS

The personnel involved in data collection can influence the quality of the data aboveall in methods where a high taxonomic knowledge is required This influencedepends on the measure to which taxonomic accuracy is needed Wilkie et al.50observed that in large-scale invertebrate biodiversity assessment, the postcollectingprocessing requires considerable effort for sorting, identification, and cataloging ofthe material The time demand on specialists to process this material would havesevere financial implications for any project.51 It is possible to reduce costs in thisphase of the project by reducing the role of the specialists A variety of strategies52,53have been developed for rapid postfield processing of data, such as sorting to highertaxonomic level only (“taxonomic sufficiency”) or employing nonspecialist techni-cians to separate specimens into informal groups based on obvious external charac-teristics (morphospecies) Oliver and Beattie54 presented the results of studies inwhich technicians were used, after a one-off training period, to separate specimensinto morphospecies Samples were sent to experts for checking and a multivariateanalysis was performed on both datasets They found that the results using thissimplified approach were very similar to those obtained using species data andconcluded that identification errors were insufficient to affect results and conclusions.The use of guilds or morphological groups as indicators for monitoring changes

in ecosystem function has been considered by several authors55,56 as a good promise between the need for specialized knowledge and rapid field proceduresemploying nonspecialist technicians In particular, McCune et al.57 observed thatrandom subsamples of species (community level) tended to produce the same pat-terns as the complete dataset If, in fact, groups of species are defined by their growthforms (for example foliose, fruticose, and crustose lichens) or their ecology (epi-phytic and epiphytic lichens), then different patterns of biodiversity and composi-tional gradients will result.58

com-Notwithstanding these results, perhaps the simplest and most readily cated descriptor of diversity is species richness This parameter, however, depends

communi-closely on the skill, the taxonomic knowledge, and the willingness of the trainee tolearn Although some trainees are strongly motivated, most of them need to feel thatbiodiversity assessment is important.57 It is obvious that improving the taxonomictraining of field observers can significantly reduce the negative bias in estimatingspecies richness.

20.7.4 OBSERVER ERROR IN LICHEN DIVERSITY MONITORING

The assessment of lichen diversity is affected by various kinds of errors, mainlyregarding the sampling procedures and the influence of environmental factors (seeprevious paragraphs) In regard to the observer error, important sources of datavariability are represented by:

• Subjectivity in the positioning of the sampling grid

• Subjectivity in the selection of trees (sampling tactic) on which the relevésare carried out

• Taxonomic error in the identification of the species in trees

• Count of the frequency of the species appearing in the grid

In order to reduce the influence of the subjectivity of the operators, Italianguidelines for assessing air pollution using lichens59 recently were changed to adopt

a new sampling grid calling for monitoring of the exposure of the whole trunk.Furthermore, standardization procedures were implemented, also in order toassess the repeatability of different sampling tactics in selecting the trees to be relevéd.The results showed that a very simple tactic is not always repeatable as it is toosubjective, while an excessively complex tactic, although it is more objective, isdifficult to apply in the field, particularly by inexperienced personnel This problemcan be solved through personnel training and harmonization procedures

A similar approach was suggested in the American guidelines for lichen diversitymonitoring,60 where only macrolichens are collected and crustose lichens are excluded

As an obvious conclusion, experienced observers found, in general, more speciesthan intermediate observers and beginners However, McCune et al.57 noted thatduring the intercalibration ring tests with operators from different regions, familiaritywith the local flora could affect the results

A method requiring a high level of taxonomic knowledge is in the Italianguidelines for monitoring effects of atmospheric pollution with lichens The samplingdesign of the protocol is based on the assessment of the frequency and abundance ofall lichen species, including groups of lichens that are difficult to identify in thefield, such as crustose lichens.59

More detailed biomonitoring data may be obtained using this approach but, onthe other hand, the sources of error due to insufficient taxonomic knowledge aremore significant and should be carefully assessed by means of intercalibration testsducted in Italy45 which confirms the influence of operator experience (taxonomicknowledge) on the results The operators were grouped into three classes based ontheir lichenological (taxonomic) experience: low, medium, and high Accuracy oftaxonomic identification was strongly affected by this parameter, ranging among theand quality control procedures Table 20.4 shows an intercalibration ring test con-

groups from 28.9% (low experience) to 61.2% (high experience) It is noteworthythat the Measurement Quality Objective (MQO) for accuracy of taxonomic identi-fication, stated at 75%, was reached only by a small number of “very experienced”operators, while the MQO for quantitative (frequency of the species) and qualitative(number of species) accuracy were only reached by the groups with average, andmuch lichenological, experience, suggesting that the average quality of the dataproduced by the operators must be improved.45

The influence of the learning factor on the results of the survey was examined

by Brunialti et al.45 who reported the results of two intercalibration tests of theoperators involved in a lichen monitoring program in Italy carried out before andafter a 7-months’ training period The operators showed a great improvement in dataquality (Table 20.5), in terms of both quantitative lichen biodiversity accuracy (from72.4 to 84.6%) and accuracy of taxonomic identification (from 32.1 to 56.1%).Furthermore, operator accuracy often improved during the same test, and the resultsshowed that accuracy improved with taxonomic training and, above all, continuousfieldwork and harmonization procedures Since the task of long-term monitoring is

to measure trends of environmental variables over time, the learning factor maycause confusion between the trend of variation of the variable of interest and the

TABLE 20.4

Percentage Accuracy in Lichen Biomonitoring Studies, in Relation

to the Level of Experience of the Operators

Level of

Lichenological

Experience

Quantitative Accuracy (%)

Qualitative Accuracy (%)

Taxonomic Accuracy (%)

Source: Adapted from Brunialti, G et al., Evaluation of data quality in lichen biomonitoring studies:

the Italian experience, Environ Monit Assess., 75, 271, 2002.45

Source: Adapted from Brunialti, G et al., Evaluation of data quality in lichen biomonitoring studies:

the Italian experience, Environ Monit Assess., 75, 271, 2002.45

trend of improvement of the crews As observed by McCune et al.,57 it is possiblethat observer error will change over a period of time with improvement of skill andchanges in motivation Another aspect to consider is also familiarity with the regionalflora As a consequence, we should also consider that perception depends on pre-conception, in that we tend to see the species we expect to see.57 To minimize thiserror, it is advisable to check particularly carefully the early phase of the survey,planning remeasuring procedures or field assistance of specialists (harmonization).For example, in the case of lichen biomonitoring, crustose lichen species are oftenoverlooked by nonspecialists, even if in many habitats they account for most of thelichens occurring in the relevé Furthermore, crustose lichens are often the pioneeringphase of colonization on trees, and in many situations they indicate a significantimprovement of air quality When considering long-term monitoring with surveysevery 2 years, if a crew, after a training period, relevés crustose lichens only in thesecond survey, these data may be interpreted as improved environmental quality.But who can say whether crustose lichens were already there also during the firstsurvey? This question can only be answered by remeasurement in the early phase

of the survey by an expert Some simple recommendations may also minimize thiskind of error For example, the American Field Method Guide60 suggests that whilecollecting the data, the crew should consider everything looking like a lichen as alichen This could result in an overestimation but could balance the underestimationcaused by the crew’s low taxonomic knowledge

20.7.5 O BSERVER E RROR IN O ZONE M ONITORING WITH

T OBACCO P LANTS

There are several biomonitoring techniques that do not require taxonomical edge or biodiversity assessment This is the case of biomonitoring of tropospheric

knowl-ozone using sensitive tobacco (Nicotiana tabacum Bel W3) plants (e.g.,

Heggestad61) With this method operators are asked to score the amount of leaf areainjured by typical ozone-induced necrosis Scoring is carried out according to acategorical rating system after a visual examination of the leaves Given the subjec-tivity of the visual estimates, differences between observers are likely to occur.Intercalibration exercises carried out in Italy show different levels of agreementamong observers Lorenzini et al.46 report a high average repeatability (65.4%) Asexpected,41 they also saw that agreement on extreme classes of injury is easy, whilecentral classes were evaluated less consistently Differences between observers may

be related to the distribution of injury on the leaf and by the extent of injury As arule, when injury covers less than 50% of total leaf area the eye would focus on thediseased tissue and vice versa.62

A series of intercalibration exercises were carried out within a more formalquality assurance program developed for a biomonitoring project in Florence, Italy.63

QA was based on established measured quality objectives (MQOs) and data qualitylimits (DQLs) MQOs allow an operator to have a ±1 category score; DQLs requireobservers to match MQOs in 90% of cases Results obtained after 4350 comparisonsbetween different observers show that a complete agreement (no difference) wasreached in 49% of cases, while deviations ±1 class were obtained in 35% of cases

(Figure 20.1) Thus, MQOs were actually matched in 84% of cases which is a goodresult, although below the desired DQLs These examples show that biomonitoringdata can be affected by observer bias Documentation of differences between observ-ers is essential in order to keep track of their performance and to allow further cross-evaluation of the data If documentation is available, unusual records may beexplained by observer bias, or, in a different perspective, systematic deviations may

be managed mathematically The need for a continuous effort toward more stringentprotocols and harmonization procedures among operators and experts is, therefore,obvious, and when adequate documentation is available, it will be possible to trackthe improvements obtained in data consistency

20.7.6 OBSERVER ERROR IN TREE CONDITION SURVEYS

The condition of the European forests has been monitored since 1986/1987 withinthe UN/ECE International Cooperative Programme on Assessment and Monitoring

of Air Pollutant Effects on Forests (ICP Forests) and the European Union Scheme

on the Protection of Forests against Atmospheric Pollution The basic assessmentmethod is a visual assessment of defoliation of a number of sample trees located atthe intersections of a nominal 16 ∞ 16 km grid In 2002, 5,929 plots with more than131,000 trees were assessed in 30 European countries.64 The survey is intended toprovide data about spatial and temporal variation of tree defoliation in Europe inrelation to air pollution and other stress factors Obviously, the potential for datacomparisons in space and time is fundamental A number of problems have been

FIGURE 20.1 Difference between observers: 0 means no difference (complete agreement);

1 means 1 class difference, etc The rating system was based on nine classes (0, 1, 2, 3, 4,

5, 6, 7, 8) Differences up to ±1 class were considered within the MQOs (From Ferretti, unpublished data.)

Difference between Observers

0 10

n = 4350

identified with the design of this program (e.g., Innes65): however, the most seriousone concerns the comparability of the assessment done by different surveyors oper-ating in different countries.66 A series of papers have addressed this issue (e.g., see

and between countries is problematic The problem is mainly rooted in differentmethods and reference standards applied by the various countries participating inthe program and can be exacerbated by peculiar species, individuals, and site con-dition.68 As these problems were not adequately addressed at the beginning of theprogram, they now have resulted in a strong impact on spatial comparability Forexample, the maps published in the UN/ECE reports now always have a captionwarning about this problem Unfortunately, a similar problem is likely to occur alsofor temporal comparisons Consider that long-term monitoring programs mayencounter considerable turnover in the personnel involved and shift and changes inthe methods applied Actually, there is some evidence that such changes haveoccurred, and this may weaken the reliability of the time series.74 In this situation,proper QA procedures are essential; recently, actions were undertaken for the design

of a series on international cross-calibration courses63 as well as to develop analysis assessment techniques (e.g., Mizoue75) These initiatives will not solve theproblem of comparability; however, they will allow proper documentation and QC

image-on the data series

20.7.7 M ANAGEMENT OF THE L EARNING F ACTOR

The learning factor — the acquisition of experience while conducting surveys —should not be underestimated, particularly in long-term monitoring This parameter

is controllable by means of remeasuring and harmonization, especially during theearly monitoring periods With this latter procedure, the experience of an expert can

be transmitted directly to the rest of the crew and offset the assessment or timation errors that could be made in the earlier stages of the work Variability due

underes-to the learning facunderes-tor may be prevented by employing for the survey only peoplewho have successfully completed training and certification and who meet the MQOrequisites However, this may be difficult and limiting in a cost-effectiveness per-spective in the case of large-scale monitoring for which a large number of operatorsare needed.42

20.7.8 T IME R EQUIRED FOR E ACH S AMPLING P HASE

In order to save financial resources when planning biomonitoring networks, it isimportant to know how much time will be required for fieldwork In order to estimatethe applicability of the different tactics, the time required for each sampling procedurewill be indicated and the main difficulties found during the fieldwork taken intoaccount The influence on data quality of the times required for each phase of thesampling protocol has been considered by several authors.32,50,57 It is obvious that themore time we spend on the sampling procedures, the better results we obtain in terms

of data quality However, we must also consider lowering the costs of the procedures

An example is provided by the issue of site access A widespread ecological surveyReferences [10, 43, 67–73]) All papers agree that comparison between observers

will also include visits to remote sites which may be difficult to reach due to time,cost, and safety concerns.17 Some monitoring informations can be obtained throughremote sensing using aerial photography, videogeography, or satellite imagery, butmany ecological variables require actual field visits In these cases, probability sam-pling can be used to reduce the number of “difficult” sites to be sampled.

Particularly for biological diversity monitoring, the time at disposal for eachplot sampling will influence the results considerably In the case of the Forest HealthMonitoring program,57 the sampling is time-constrained to help standardize effortacross crews and to facilitate scheduling crew activities

20.8 CONCLUSIONS

Biomonitoring is a powerful approach in various environmental studies However,biomonitoring investigations are subjected to a variety of error sources that need to beacknowledged and documented in order to be managed properly Error sources encom-pass a variety of subjects, from the design of the investigation to indicators developmentand data collection As decision makers need robust and defensible data, environmentalscientists need to consider proper procedures for reassurance about the value of theirdata QA is essential in this respect as it allows the identification of the various errorsources and forces investigators to address and solve problems It is therefore importantthat environmental biologists and field ecologists consider QA as a key attribute oftheir work In agreement with References 1, 5, and 7, we suggest that QA and relatedsubjects should be considered an integral part of environmental surveys

Various examples support our suggestion: biomonitoring surveys using lichens,specific indicator plants, and spontaneous vegetation have reported problems in datacomparability at a variety of spatial and temporal scales While a series of inherentproperties of the biological systems will make it almost impossible to achieve a fullcomparability of biomonitoring data, even the simple documentation of the errorsassociated with the various investigation steps will provide many benefits In par-ticular, it will allow investigators, end-users, and the public to know the confidence

to be placed on the data As environmental studies are almost always based on fundsfrom public agencies, we think that the extent to which the quality of the data can

be documented should be a criterion that may help in distinguishing the value ofbiomonitoring (and other) studies

REFERENCES

1 Wagner, G., Basic approaches and methods for quality assurance and quality control

in sample collection and storage for environmental monitoring, Sci Total Environ.,

264, 3, 1995.

2 Ferretti, M and Erhardt, W., Key issues in designing biomonitoring programmes.

Monitoring scenarios, sampling strategies and quality assurance, in Monitoring with Lichens — Monitoring Lichens, Nimis, P.L., Scheidegger, C., and Wolseley, P., Eds.,

Kluwer Academic, Dordrecht, Netherlands, 2002, chap 9.

3 Kovacs, M., Biological Indicators in Environmental Protection, Horwood, New York,

1992, 207 pp

4 Gertner, G et al., Mapping and uncertainty of predictions based on multiple primary variables from joint co-simulation with Landsat TM image and polynomial regression,

Remote Sens Environ., 83, 498, 2002.

5 Kohl, M., Traub, B., and Paivinen R., Harmonisation and standardisation in

multi-national environmental statistics — mission impossible?, Environ Monit Assess., 63,

361, 2000.

6 Cline, S.P and Burkman, W.G., The role of quality assurance in ecological programs,

in Air Pollution and Forest Decline, Bucher, J.B and Bucher-Wallin, J., Eds., IUFRO,

Nadel-men, Allg Forst Jagdztg., 162, 210, 1991.

11 Cochran, W.G., Sampling Techniques, 3rd ed., John Wiley & Sons, New York, 1977,

643 pp.

12 Kish, L., Survey sampling, John Wiley & Sons, New York, 1965, 643 pp.

13 Gertner, G and Kohl, M., An assessment of some non-sampling errors in national

survey using an error budget, Forest Sci., 41(4), 758, 1992.

14 Lessler, J and Kalsbeek, W., Nonsampling Error in Surveys, John Wiley & Sons,

New York, 1992, 412 pp.

15 Yoccoz, N.G., Nichols, J.D., and Boulinier, T., Monitoring of biological diversity in

space and time, Trends Ecol Evol., 16, 446, 2001.

16 Laskowsky, S.L and Kutz, F.W., Environmental data in decision making in EPA

regional offices, Environ Monit Assess., 51, 15, 1998.

17 Olsen, A.R et al., Statistical issues for monitoring ecological and natural resources

in the United States, Environ Monit Assess., 54, 1, 1999.

18 Will-Wolf, S and Scheidegger, C., Monitoring lichen diversity and ecosystem

func-tion, in Monitoring with Lichens — Monitoring Lichens, Nimis, P.L., Scheidegger, C., and Wolseley, P., Eds., Kluwer Academic, Dordrecht, Netherlands, 10, 2002.

19 Giordani, P., Brunialti, G., and Modenesi, P., Applicability of the lichen biodiversity

method (L.B.) to a Mediterranean area (Liguria, NW Italy), Cryptogamie Mycol., 22,

193, 2001.

20 Brunialti, G and Giordani, P., Variability of lichen diversity in a climatically

heter-ogeneous area (Liguria, NW Italy), Lichenologist, 35, 55, 2003.

21 Loppi, S et al., A new scale for the interpretation of lichen biodiversity values in the

Thyrrenian side of Italy, in Progress and Problems in Lichenology at the Turn of the Millennium, 4th IAL Symposium (IAL 4), Llimona, X., Lumbsh, H.T., and Ott, S., Eds., Bibliotheca Lichenologica, 82, Gebruder Borntraeger Verlagsbuchhandlung,

Stuttgart, 2002, 235.

22 Guisan, A and Zimmermann, N.E., Predictive habitat distribution models in ecology,

Ecol Model., 135, 147, 2000.

23 Loppi, S et al., Identifying deviation from naturality of lichen diversity in

heteroge-neous situations, in Monitoring with Lichens — Monitoring Lichens, Nimis, P.L.,

Scheidegger, C., and Wolseley, P., Eds., Kluwer Academic, Dordrecht, Netherlands,

2002, 281.

24 Nimis, P.L and Martellos, S., Testing the predictivity of ecological indicator values:

A comparison of real and ‘virtual’ relevés of lichen vegetation, Plant Ecol., 157, 165,

2001.

25 Nimis, P.L et al., Biomonitoring of trace elements with lichens in Veneto (NE Italy),

Sci Total Environ., 255, 97, 2000.

26 Bargagli, R., Trace Elements in Terrestrial Plants: An Ecophysiological Approach to Biomonitoring and Biorecovery, Springer-Verlag & R.G Landes, Berlin, 1998, chap.

8 and 9.

27 Bargagli, R and Mikhailova, I., Accumulation of inorganic contaminants, in toring with Lichens — Monitoring Lichens, Nimis, P.L., Scheidegger, C., and Wolse- ley, P., Eds., Kluwer Academic, Dordrecht, Netherlands, 2002, chap 6.

Moni-28 Minganti, V et al., Biomonitoring of trace metals by different species of lichens

(Parmelia) in North-West Italy, J Atmos Chem., 45, 219, 2003.

29 Nimis, P.L., Andreussi, S., and Pittao, E., The performance of two lichen species as

bioaccumulators of trace metals, Sci Total Environ., 275, 43, 2001.

30 Houston, L et al., A multi-agency comparison of aquatic macroinvertebrate-based

stream bioassessment methodologies, Ecol Indicat., 1, 279, 2002.

31 Knopman, D.S and Voss, C.I., Multi-objective sampling design for parameter

esti-mation and model discrimination in groundwater solute transport, Water Resour Res.,

25, 2245, 1989.

32 McCune, B and Lesica, P., The trade-off between species capture and quantitative accuracy in ecological inventory of lichens and bryophytes in forests in Montana,

Bryologist, 95, 296, 1992.

33 Humphrey, J.W et al., The importance of conifers plantations in northern Britain as

a habitat for native fungi, Biol Conserv., 96, 241, 2000.

34 Vitt, D.H., Yenhung, L., and Belland, R.J., Patterns of bryophyte diversity in peatlands

of western Canada, Bryologist, 98, 218, 1995.

35 Collins, S.L and Glenn, S.M., Effects of organismal and distance scaling on analysis

of species distribution and abundance, Ecol Appl., 7, 543, 1997.

36 Qian, H., Klinka, K., and Song, X., Cryptograms on decaying wood in old-growth

forests of southern coastal British Columbia, J Veg Sci., 10, 883, 1999.

37 Rose, F., Ancient British woodlands and their epiphytes, Br Wildl., 5, 83, 1993.

38 Ferretti, M et al., Reliability of different sampling densities for estimating and

mapping lichen diversity in biomonitoring studies, Environ Pollut., 127, 249, 2004.

39 Hunsaker, C.T., New concepts in environmental monitoring: the question of

indica-tors, Sci Total Environ., Suppl., 77, 1993.

40 Breckenridge, R.P., Kepner, W.G., and Mouat, D.A., A process to select indicators

for monitoring of rangeland health, Environ Monit Assess., 36, 45, 1995.

41 Muir, P.S and Mc Cune, B., Index construction for foliar symptoms of air pollution

injury, Plant Dis., 71, 558, 1987.

42 Tegler, B., Sharp, M., and Johnson, M.A., Ecological monitoring and assessment network’s proposed core monitoring variables: an early warning of environmental

change, Environ Monit Assess., 67, 29, 2001.

43 Innes, J.L., Landmann, G., and Mettendorf, B., Consistency of observations of

defo-liation amongst three different European countries, Environ Monit Assess., 25, 29,

1993.

44 Ferretti, M., Potential and limitations of visual indices of tree condition, sphere, 36, 1031, 1998.

Chemo-45 Brunialti, G et al., Evaluation of data quality in lichen biomonitoring studies: the

Italian experience, Environ Monit Assess., 75, 271, 2002.

46 Lorenzini, G et al., Visual assessment of foliar injury induced by ozone on indicator

tobacco plants: a data quality evaluation, Environ Monit Assess., 62, 175, 2000.

47 Barker, J.R et al., Evaluation of metric precision for a riparian forest survey, Environ Monit Assess., 75, 51, 2002.

48 McRoberts, R.E et al., Variation in forest inventory field measurements, Can J For Res., 24, 1766, 1994.

49 Englund, S.R., O’Brien, J.J., and Clark, D.B., Evaluation of digital and film spherical photography and spherical densiometry for measuring forest light environ-

hemi-ments, Can J For Res., 30, 1999, 2000.

50 Wilkie, L., Cassis, G., and Gray, M., A quality control protocol for terrestrial

inver-tebrate biodiversity assessment, Biodivers Conserv., 12, 121, 2003.

51 Marshall, S.A et al., Terrestrial arthropod biodiversity: planning a study and

recom-mended sampling techniques, Bull Entomol Soc Can., 26, 1, 1994.

52 Oliver, I and Beattie, A.J., Invertebrate morphospecies as surrogates for species: a

case study, Conserv Biol., 10, 99, 1996.

53 Beattie, A.J and Oliver, I., Taxonomic minimalism, Trends Ecol Evol., 9, 488, 1994.

54 Oliver, I and Beattie, A.J., A possible method for the rapid assessment of biodiversity,

Conserv Biol., 7, 562, 1993.

55 Will-Wolf, S., Esseen, P.-A., and Neitlich, P., Monitoring biodiversity and ecosystem

function: forests, in Monitoring with Lichens—Monitoring Lichens, Nimis, P.L., Scheidegger, C., Wolseley, P., Eds., Kluwer Academic, Dordrecht, Netherlands, 2002,

chap 14.

56 Will-Wolf, S., Scheidegger, C., and McCune, B., Methods for monitoring

biodi-versity and ecosystem function, in Monitoring with Lichens — Monitoring Lichens, Nimis, P.L., Scheidegger, C., and Wolseley, P., Eds., Kluwer Academic, Dordrecht,

Netherlands, 2002, p 11.

57 McCune, B et al., Repeatability of community data: species richness versus gradient

scores in large-scale lichen studies, Bryologist, 100, 40, 1997.

58 McCune, B and Antos, J.A., Correlations between forest layers in the Swan Valley,

61 Heggestad, H.E., Origin of Bel W3, Bel C and Bel B tobacco varieties and their use

as indicators of ozone, Environ Pollut., 74, 264,1991.

62 Horsfall, J.G and Cowling, E.B., Pathometry: the measurement of plant disease, in

Plant Disease: An Advanced Treatise, Vol II, Academic Press, New York, 1978.

63 Ferretti, M et al., A biomonitoring network to upscale site-related measurements of

ground level ozone in the area of Florence, Italy, EuroBionet2002 Conference on Urban Air Pollution, Bioindication and Environmental Awareness, University of

Hohenheim, November 5–6, Summary 21, 2002.

64 Lorenz, M et al., Forest Condition in Europe: Results of the 2002 Large-scale Survey,

UN/ECE and EC, Geneva, 116 pp plus Annexes, in press.

65 Innes, J L., Forest health surveys — a critique, Environ Pollut., 54, 1, 1988.

66 Cozzi, A., Ferretti, M., and Lorenz, M., Quality Assurance for Crown Condition

Assessment in Europe, UN/ECE, Geneva, 2002, 111 pp.

67 Neumann, M and Stowasser, S., Waldzustandinventur: zur Objektivitat von

Kronen-klassifizierung, in Fortsliche Bundesversuchsanstalt Wien, Jahresbericht 1986,

Vienna: fortsliche Bundesversuchsanstalt, 1986, pp 101–108.

68 Innes, J L., Forest health surveys: problems in assessing observer objectivity, Can.

J For Res., 18, 560, 1988.

69 Schadauer, K., Die Ermittlung von Genauigkeitsmassen terrestrischer zustandsinventuren im Rahmen der ostereichischen Waldzustandsinventur, 1991.

Kronen-70 Köhl, M., Quantifizierung der Beobachterfehler bei der Nadel-/Blattverluschatzung,

Allg Forst Jagdztg., 164, 83, 1992.

71 Ghosh, S., Innes, J.L., and Hoffman, C., Observer variation as a source of error in

assessment of crown condition through time, Forest Sci., 41, 235, 1995.

72 Ferretti, M., Cenni, E., and Cozzi, A., Indagini sulle condizioni dei boschi Coerenza

econfrontabilità dei dati sulla trasparenza delle chiome degli alberi in Italia, Montie Boschi, 2, 5, 1994.

73 Ferretti, M., Cenni, E., and Cozzi, A., Assessment of tree crown transparency and crown discoloration as performed by surveyors from six European countries during the 7th Intercountry EC/ECE Mediterranean Intercalibration Course, Cagliari, Italy,

1994, Regione Autonoma della Sardegna, 1995, 32 pp plus Annexes.

74 Landmann, G., Nageleisen, M., and Ulrich, E., New evidence for a methodological

shift in the visual assessment of the crown condition of broadleaves, Cah DSF, 1,

74, 1998.

75 Mizoue, N., CROCO: semiautomatic image analysis system for crown condition

assessment in forest health monitoring, J For Plann., 8, 17, 2002.

Patchy Distribution Fields: Acoustic Survey Design and Reconstruction Adequacy

I Kalikhman

CONTENTS

21.1 Introduction 46521.2 Mathematical Model Design 46721.3 Mathematical Experiments 46821.4 Conclusions 496Acknowledgments 497References 497

21.1 INTRODUCTION

Patchiness is a fundamental attribute of ecosystems.1 Each ecosystem componenthas its typical dimension of patches The scale of the patchiness for fish concentra-tions ranges from several feet to hundreds of miles.2 For plankton, it is usuallysmaller than that for fish and ranges from one foot to several miles or even dozens

of miles.3 Gaps are a special case of patchiness, corresponding to areas of habitablespace in which organisms are noticeably reduced in abundance relative to back-ground levels.4 Because of the patchiness of ecosystem component distribution, theparameters of an acoustic survey should be chosen on the basis of statistical char-acteristics of such fields

One major goal of any acoustic survey is to reconstruct the distribution fieldstudied In this case, the survey design is considered efficient enough if it ensuresthe required adequacy of a reconstructed field to the actual one with minimalexpenditures.5 It is assumed in this consideration that the size and structure of theactual distribution field, under natural conditions, should be known However, in areal situation, the parameters of distribution fields are never known Therefore, themathematical simulation method is used to determine the efficiency of survey designand improve the algorithm of data analysis

21

L1641_C21.fm Page 465 Tuesday, March 23, 2004 7:40 PM

466 Environmental Monitoring

The efficiency of an acoustic survey mainly depends on its pattern, the distancebetween transects, and the unit of sampling distance The survey pattern representseither zigzag or parallel transects.6 Zigzag transects are the more economical surveypattern In the case of zigzag transects, the survey path is formed entirely by transects;for parallel ones, it consists of transects and connecting tracks Parallel transects aremainly used only if the borders of the region are known.7 In this case, the collection

of data is possible not only on transects but also on connecting tracks The position

of the starting point may, to some extent, affect the efficiency of surveys

The choice of the distance between transects usually depends on the surveyscale In carrying out large-scale surveys, the distance between transects may reach

50 to 60 mi or more; with small-scale surveys, it may be from 20–30 down to 4–5 mi

or less It was recommended to locate transects at a distance close to the autocorrelationradius for the field.7 However, no relationships for determining the accuracy of surveyswere established In studies aimed at the revelation of these relationships, the relativeerror of the abundance estimate (the bias of the survey) was used as a measure of thesurvey quality.8,9

The unit of sampling distance is the length of a survey transect along whichthe acoustic measurements are averaged to receive one sample If the samplingdistance unit is too large, potentially useful information about the distribution ofthe stock will be lost If it is too small, successive samples will be correlated, inwhich case it will be difficult to determine the confidence interval for the stockestimate.6 The unit of sampling distance may be as short as 0.1 mi to distinguishdense schools or as long as 10 mi in the case of species widely distributed overlarge areas of the ocean; usually, the sampling distance unit may be within therange of 1 to 5 mi.6

The aim of the present study is to determine survey design parameters (thesurvey pattern, the distance between transects, and the unit of sampling distance),allowing one to obtain a realistic image of a patchy distribution field The bias ofacoustic surveys is important in studies aimed at estimation of the fish abundance.However, since acoustic surveys began to be used for ecological purposes (e.g.,Reference 10), it is often necessary to estimate interrelationships in the organization

of populations without considering the bias Therefore, herein, survey design eters are determined allowing the unbiased reconstruction of an original distributionfield This means that the bias in the survey results is neglected; the parameters ofthe reconstructed field and of that originally generated are compared by correlationanalysis Fish and zooplankton distribution in Lake Kinneret (Israel) are used asprototypes in simulating patchy fields

param-If the efficiencies of various survey patterns are compared under the assumption

of fixed transect spacing, it means that the numbers of parallel or zigzag transectsare equal to each other However, as the length of a zigzag transect is always smallerthan that of a parallel transect plus connecting track, the sampling efforts (theoverall survey path or, in other words, the time for conducting a survey) for bothpatterns are not the same Therefore, the efficiencies of various patterns are oftencompared under the assumption of fixed sampling effort, when the overall surveypaths for both patterns are equal to each other In this case, a greater number ofzigzag transects compared to parallel ones may be carried out As a result, the

L1641_C21.fm Page 466 Tuesday, March 23, 2004 7:40 PM

Patchy Distribution Fields: Acoustic Survey Design 467

adequacy of reconstructing a field from data of a zigzag survey is expected toincrease; determination of the extent of this increase is one of the objectives of thepresent study

21.2 MATHEMATICAL MODEL DESIGN

The mathematical model is based on the approach described in References 11 and

12 The surveyed area is a rectangular array of numbers making up a square matrix

of 50 lines and 50 columns Each of these 2500 numbers represents the value of afield in an elementary area (node) The distance between the nodes is equal to one,

so the grid size is 50 × 50 To simulate a distribution field, patches (or gaps) aregenerated and placed into the surveyed area Outside patches the field variable iszero, while outside gaps it is the constant background Inside a patch (gap) thevariable increases (decreases) from the outer border towards the center The proba-bilistic distribution low of the appearance of various field values in a patch (gap) isPoisson’s (or the logarithmical) normal; this is in agreement with the results obtained

by analyzing data from real acoustic microsurveys.13

Patches (gaps) may be isotropic (circular or nondirectional) or anisotropic tical or directional) with various sizes and spatial orientation; or separate or over-lapping (partially or fully), forming larger patches (gaps) with more complicatedshapes Patches (gaps) are either immovable (static) or movable (dynamic); movingpatches (gaps) have a realistic shape like a comet.2 Random fluctuations in densitycan be superimposed on the simulated patchy field

(ellip-A resulting (constructed) patchy distribution field is then sampled by simulating

an acoustic survey Thestarting point (the beginning of the initial transect) is alwayslocated either on the abscissa or ordinate axis We distinguish the general direction

of a survey and the survey path The survey direction coincides either with the abscissa

or ordinate axis The survey path with an overall length S is formed by transectsdisposed as parallel lines or a zigzag Transects, as parallel lines, are disposedperpendicularly to the general survey direction Thus, the overall survey path in thiscase consists of parallel transects and connecting tracks perpendicular to transects.The distances between parallel transects are regular and equal to the length of con-necting tracks (D)

A zigzag is characterized by the shift between neighboring parallel transects inthe general survey direction The shift is a move of a survey when passing from onetransect to another The shift between neighboring parallel transects is regular andtwice as large as the distance between corresponding parallel transects Throughoutthis study, half a shift between neighboring parallel transects is called the distancebetween zigzag transects (D) Along transects, a unit of sampling distance can be set

As mentioned above, this distance is the length of a transect over which the values

of a field are integrated and averaged In the model, the sampling distance unit (d)

is equal to a specific part of the length of a transect (the minimal value d = 1/50).The values of a field along the survey path over each sampling distance areconsidered to be measured without error and are used to reconstruct a field Thereconstructed field consists again of the full set of nodes of the array representingthe surveyed area The corresponding values of the reconstructed field and that

L1641_C21.fm Page 467 Tuesday, March 23, 2004 7:40 PM

468 Environmental Monitoring

originally generated are then compared for all nodes Their adequacy is evaluated

by the coefficient of determination representing the square of the standard Pearsoncorrelation coefficient (r).14

Further on, the reconstruction of distribution fields will be taken into ation The gridding methods use weighted interpolation algorithms The krigingmethod widely used in reconstructing distribution fields is tested This algorithmassumes an underlying variogram, which is a measure of how quickly things change

consider-on the average.15 The method of gridding is chosen so that the maximal correlationbetween the reconstructed field and that originally generated is attained Preliminarysimulations indicate that in most cases, the maximal correlation is attained whenusing kriging with the linear variogram model Application of various griddingmethods may have an impact on relationships of interest; therefore, to exclude thisimpact, the indicated method is used alone Search option controls which data pointsare considered by the gridding operation when interpolating grid nodes.15

The autocorrelation radius is used as a characteristic of a field in choosing theparameters of a survey.16 The autocorrelation radius (R) is determined from a plot

of the autocorrelation function as the distance from the coordinate center to the pointwhere the lag (autocorrelation) is zero An important property of the autocorrelationfunction consists of the fact that it is sensitive only to variable deviations, whileconstant levels have no effect.17 This enables us to use the autocorrelation function

in seeking the autocorrelation radii of distribution fields that include gaps out the study, R, S, and D are given in proportion to the size of the square representing

Through-a surveyed Through-areThrough-a

21.3 MATHEMATICAL EXPERIMENTS

In this section, the results of simulations reflecting the reconstruction of static anddynamic distribution fields are considered

where: the transect spacing is fixed, a given sampling effort is allocated to a survey,and a choice as made for the unit of sampling distance

Fixed transect spacing. Let us consider the set of problems in application to theparallel survey pattern Examples of the original isotropic field are given in

size of the patches and is equal to R = 0.15 (left), R = 0.09 (middle), or R = 0.10(right), independent of the direction chosen (second row).18 The adequacy of recon-structing distribution fields is also independent of the direction of the survey per-formed For example, the simulations show that the surveys carried out at D = 0.20

in two perpendicular directions give similar results of r2= 0.81 and r2= 0.85 (left),

r2= 0.20 and r2= 0.28 (middle), or r2= 0.40 and r2= 0.38 (right) (the number ofsample data points NSDP = 300) The increase of the distance between transects(compare the third row with fourth) and the decrease of the autocorrelation radius(compare the left column with the middle and right columns) both lead to lowercorrelation between the reconstructed field and that originally generated Thus, theconclusion can be made that the adequacy of a reconstructed field to its originaldepends upon the D/R ratio

L1641_C21.fm Page 468 Tuesday, March 23, 2004 7:40 PM

Figure 21.1, first row The autocorrelation radius for such a field corresponds to the

Patchy Distribution Fields: Acoustic Survey Design 469

FIGURE 21.1 First row: Original patchy isotropic fields Second row: Autocorrelation circles for the fields Third and fourth rows: Paths of the simulated surveys and the reconstructed fields Right column represents the gaps (the isolines are labeled by negative numbers).

L1641_C21.fm Page 469 Tuesday, March 23, 2004 7:40 PM

470 Environmental Monitoring

Original anisotropic fields are generated as a set of anisotropic patches with similar

anisotropy ratio (AR) and anisotropy angle (AA).15 Autocorrelation radii in directions

of abscissa and ordinate axes are equal to RX = 0.27 and RY = 0.13 (AR = 4, AA = 0°,

left), RX = 0.06 and RY = 0.23 (AR = 6, AA = 90°,middle), or RX= 0.25 and RY = 0.06

(AR = 6, AA = 0°, right) (second row) Simulations conducted indicate that the

surveys in the direction of patch elongation permit us to reconstruct the original

field more precisely than those in a perpendicular direction For example, a survey

carried out at D = 0.40 in the direction of patch elongation gives better results in

terms of the original field adequacy (r2= 0.92, left; r2= 0.76, middle; or r2= 0.79,

right) (NSDP = 200, third row) than even a survey conducted with the smaller

distance between transects (D = 0.20, left; D = 0.22, middle;orD = 0.16,right) in

a perpendicular direction (r2 = 0.66, NSDP = 300, left; r2= 0.16, NSDP = 300, middle;

or r2= 0.26, NSDP = 400, right) (fourth row) The conclusion can be made that the

direction of the patch elongation is the optimal direction of an acoustic survey

The distribution fields were rotated to a certain angle in order to examine the

efficiency of surveys which, for some reason, cannot be conducted in the optimal

direction For each rotated original anisotropic fields (an example of rotation for 30°

and ordinate axes are equal to RX = 0.25 and RY = 0.18 (AR = 4, AA = 30°,left), RX

= 0.11 and RY = 0.20 (AR = 6, AA = 120°,middle), or RX = 0.24 and RY = 0.16 (AR

= 6, AA = 150°, right) In the case of the left field, the survey carried out at the

distance between transects D = 0.40 with the angle between the survey direction and

that of patch elongation equaling 60° gives poorer results in terms of the original

field adequacy (r2 = 0.69, NSDP = 200, fourth row) than the survey with the angle

between the survey direction and that of patch elongation equaling 30° (r2= 0.85,

NSDP = 200, third row) For the middle field, the survey carried out at D = 0.40 with

the angle between the survey direction and that of patch elongation equaling 60°

permits us to reconstruct the original field with lower adequacy (r2= 0.25, NSDP =

200, fourth row) than the survey with the angle between the survey direction and that

of patch elongation equaling 30° (r2= 0.72, NSDP = 200, third row) Finally, for the

right field, the survey carried out at D = 0.40 with the angle between the survey

direction and that of patch elongation equaling 60° gives worse results in terms of

the original field adequacy (r2= 0.66, NSDP = 200, fourth row) than the survey with

the angle between the survey direction and that of patch elongation equaling 30°

(r2= 0.80, NSDP = 200, third row) Thus, the results of experiments indicate that a

decrease in the angle between the survey direction and that of patch elongation leads

to the higher coefficient of determination between the reconstructed field and that

originally generated This is explained by the fact that D/R ratio in the direction of

patch elongation is smaller than that in any other direction or, ultimately, by the

relationship of the autocorrelation radii located in two reciprocally perpendicular

directions (the autocorrelation radius situated at a smaller angle to the major axis of

autocorrelation ellipse is always larger than that located at a larger angle to it)

Let us consider the same set of problems in application to the zigzag survey

L1641_C21.fm Page 470 Tuesday, March 23, 2004 7:40 PM

spatial orientation (Figure 21.2, first row) Anisotropic fields are characterized by the

is given in Figure 21.3, first row), the autocorrelation radii in directions of abscissa

pattern The examples of isotropic original fields, given in Figure 21.1, first row, are

reproduced in Figure 21.4, first row In the case of the left field, the surveys carried

FIGURE 21.2 First row: Original patchy anisotropic fields Second row: Autocorrelation

ellipses for the fields Third and fourth rows: Paths of the simulated surveys conducted in the

directions of major or minor axes of autocorrelation ellipses and the reconstructed fields.

Right column represents the gaps (the isolines are labeled by negative numbers).

FIGURE 21.3 First row: Original patchy rotated anisotropic fields Second row:

Autocorre-lation ellipses for the fields Third and fourth rows: Paths of the simulated surveys conducted

in the arbitrary directions and the reconstructed fields Right column represents the gaps (the

isolines are labeled by negative numbers).

FIGURE 21.4 First row: Original patchy isotropic fields Second through fourth rows: Paths

of the simulated surveys and the reconstructed fields Right column represents the gaps (the

isolines are labeled by negative numbers).

out at the distance between transects D = 0.26 in two perpendicular directions givesimilar results of r2= 0.67 (second row) and r2= 0.71 (third row) (NSDP = 198).For the middle field, the surveys conducted at D = 0.10 in two perpendiculardirections give similar results of r2= 0.71 (second row) and r2= 0.77 (third row)(NSDP = 499) For the right field, the surveys carried out at D = 0.10 in twoperpendicular directions give similar results of r2= 0.83 (second row) and r2= 0.88(third row) (NSDP = 499) Thus, regarding zigzag transects, as well as in the case

of parallel transects, the adequacy of reconstructing an isotropic field to that nally generated is actually independent on the direction of the survey performed

origi-As shown by the simulations, the decrease of the ratio of the distance betweentransects to autocorrelation radius for the field (D/R) leads to the higher coefficient

of determination between the reconstructed field and that originally generated, andvice versa

To receive an objective comparison of the results of surveys by zigzag andparallel patterns throughout the paper, not only the transect spacing is taken equalbut also their location (each parallel transect crosses the point where the middle

of a corresponding zigzag transect is situated) A comparison shows that thezigzag patterns lead to poorer results than the parallel patterns do (r2= 0.71 and

r2= 0.81, left; r2= 0.77 and r2= 0.81, middle; r2= 0.88 and r2= 0.90, right) (thirdand fourth rows)

out at the distance between transects D = 0.28 in the direction of patch elongationgives even better result in terms of the original field adequacy (r2 = 0.90, NSDP =180) (third row) than a survey conducted at D = 0.16 in a perpendicular direction(r2 = 0.64) (second row) For the middle field, a survey carried out at D = 0.34 inthe direction of patch elongation gives even better result in terms of the originalfield adequacy (r2 = 0.86, NSDP = 153) (third row) than a survey conducted at D =0.10 in a perpendicular direction (r2 = 0.67) (second row) For the right field, a surveycarried out at D = 0.48 in the direction of patch elongation permits us to reconstructthe original field much better (r2 = 0.72, NSDP = 114) (third row) than a surveyconducted at D = 0.12 in a perpendicular direction (r2 = 0.45) (second row) Thus,the simulations conducted indicate that the surveys in the direction of patch elon-gation permit us to reconstruct the original field more precisely than those in aperpendicular direction A comparison of the surveys by zigzag transects (third rows)with those by parallel transects (fourth rows) shows that the former lead to poorerresults than the latter do (r2 = 0.90 and r2 = 0.96, left; r2 = 0.86 and r2 = 0.92, middle;

r2 = 0.72 and r2 = 0.79, right)

carried out at the distance between transects D = 0.28 with the angle between thesurvey direction and that of patch elongation equaling 60° gives poorer results interms of the original field adequacy (r2 = 0.64, second row) than the survey conducted

at D = 0.34 with the angle between the survey direction and that of patch elongationequaling 30° (r2 = 0.88, NSDP = 154, third row) For the middle field, the surveycarried out at D = 0.30 with the angle between the survey direction and that of

The examples of original anisotropic fields, given in Figure 21.2, first row, arereproduced in Figure 21.5, first row In the case of the left field, a survey carried

The examples of rotated original anisotropic fields, given in Figure 21.3, firstrow, are reproduced in Figure 21.6, first row In the case of the left field, the survey

FIGURE 21.5 First row: Original patchy anisotropic fields Second through fourth rows:

Paths of the simulated surveys conducted in the directions of major or minor axes of

auto-correlation ellipses and the reconstructed fields Right column represents the gaps (the isolines

are labeled by negative numbers).

FIGURE 21.6 First row: Original patchy rotated anisotropic fields Second through fourth

rows: Paths of the simulated surveys conducted in the arbitrary directions and the reconstructed fields Right column represents the gaps (the isolines are labeled by negative numbers).

patch elongation equaling 60° permits us to reconstruct the original field with loweradequacy (r2 = 0.64, second row) than the survey conducted at D = 0.32 with theangle between the survey direction and that of patch elongation equaling 30° (r2=0.83, NSDP = 160, third row) Finally, for the right field, the survey carried out at

D = 0.22 with the angle between the survey direction and that of patch elongationequaling 60° gives worse results in terms of the original field adequacy (r2 = 0.83,second row) than the survey conducted at D = 0.28 with the angle between thesurvey direction and that of patch elongation equaling 30° (r2 = 0.88, NSDP = 179,third row) Thus, the results of experiments indicate that a decrease in the anglebetween the survey direction and that of patch elongation leads to the highercoefficient of determination between the reconstructed field and that originallygenerated The explanation is similar to that given in application to the parallelsurvey pattern A comparison of the surveys by zigzag transects (third rows) withthose by parallel transects (fourth rows) shows that the former lead to poorer resultsthan the latter do (r2= 0.88 and r2= 0.92, left; r2= 0.83 and r2= 0.90, middle; r2= 0.88and r2= 0.96, right)

The results of simulated surveys with various ratios of the distance betweenThe existence of a distinct relationship fitting the generalized dataset (r2 vs D/R)confirms the possibility of using the autocorrelation radius as a parameter of a fieldwhen choosing the distance between transects In this figure, all the range of possiblevalues for coefficient of determination is conditionally divided into an area of highvalues (>0.70) and an area of low ones (<0.70) With the D/R ratio increasing tothe critical value, the coefficient of determination remains in the first area, whilewith further increase, this coefficient falls into the second one The critical value ofthe D/R ratio depends on the survey pattern: D/R = 1.5 to 2.0, regarding the parallelpattern; D/R = 1.0 to1.5, in respect to the zigzag one For this reason, we suggestchoosing the distance between transects from the condition that the D/R ratio isequal to a critical value This distance ensures the match not less than r2 > 0.70between the field reconstructed on the basis of the data of the survey designed andthat really existing in the water body, although unknown

As in Figure 21.7, the regression curve regarding zigzag transects is locatedeverywhere below that for parallel transects; the former allow less adequate recon-struction of a field than the latter do A result of a specific mathematical experimentregarding zigzag transects may be above that for parallel transects This indicatesthat there is a great probability that the conclusion is correct

Fixed sampling effort In all of the following examples, the overall survey paths

regarding the parallel and zigzag patterns are equal to each other To demonstrate thebest results regarding the parallel or zigzag pattern, each example given below is selectedfrom the entirety of surveys with the same overall path but various starting points

overall survey path (compare the surveys presented on the left column with those given on the middle and right ones) A comparison of the surveys by parallel transects

(second row) with those by zigzag transects (third row) shows that the former lead

to poorer results than the latter do: r2 = 0.66 and r2 = 0.72 (NSDP = 200, left column),

transects to autocorrelation radius for the field (D/R) are considered in Figure 21.7

The examples of original isotropic fields, given in Figure 21.1, first row, are

reproduced in Figure 21.8, first row The larger the patches, the smaller the required

r2= 0.72 and r2= 0.81 (NSDP = 400, middle column), or r2= 0.71 and r2= 0.74

(NSDP = 400, right column)

The examples of original anisotropic fields, given in Figure 21.2, first row, are

and right fields is the abscissa, while for the middle one it is the ordinate; as shown

earlier, these are the optimal directions for acoustic survey A comparison of the

surveys by parallel transects (second row) with those by zigzag transects (third row)

shows that the former lead to poorer results than the latter do: r2= 0.91 and r2= 0.92

(left column), r2= 0.89 and r2= 0.92 (middle column), or r2= 0.80 and r2= 0.82

(right column) (NSDP = 200)

The examples of rotated original anisotropic fields, given in Figure 21.3, first

parallel transects (second row) with those by zigzag transects (third row) shows

that the former lead to poorer results than the latter do: r2= 0.86 and r2 = 0.89

FIGURE 21.7 Coefficient of determination between the reconstructed field and that originally

generated as a function of the D/R ratio with regard to various survey patterns (left: parallel;

right: zigzag) Filled symbols correspond to surveys of patchy distribution fields; empty

symbols: gappy fields Colors correspond to the results of simulated surveys of the fields

green: Figure 21.3 and Figure 21.6, middle and right (surveys corresponding to the empty

coefficient of determination regarding isotropic fields is obtained by averaging the results of

the surveys carried out in the directions of abscissa and ordinate axes The solid line represents

the nonlinear regression ensuring the minimal sum of squared residuals (the ends of the

straight lines indicate the bend of the nonlinear regression) The dashed lines are the borders

of the confidence interval with the probability of p = 0.99 The thick dashed line corresponds

to the regression for the survey pattern shown in the opposite position of the figure (left:

shown in Figure 21.1 to Figure 21.6 Dark blue: Figure 21.1 and Figure 21.4, left; red: Figure

21.1 and Figure 21.4, middle and right; crimson: Figure 21.2 and Figure 21.5, left; light blue:

dark blue, crimson, and yellow circles are not shown in Figure 21.1 to Figure 21.6) The

Figure 21.2 and Figure 21.5, middle and right; yellow: Figure 21.3 and Figure 21.6, left;

reproduced in Figure 21.9, first row The direction of patch elongation for the left

row, are reproduced in Figure 21.10, first row A comparison of the surveys by

zigzag; right: parallel) (See color insert following page 490 )

(left column), r2= 0.78 and r2= 0.87 (middle column), or r2= 0.70 and r2= 0.73(right column) (NSDP = 200).

As the path of the survey crosses the field in various directions, the averagevalue of the autocorrelation radius should be determined Because further along it

is intended to use the ratio of S/Rav to average the ratio and to represent the results

in logarithmic scale, the use of the geometric mean is preferable.14 Regarding thezigzag pattern, crossing of a field by a transect, oriented in an arbitrary direction,results in appearance of the two components of the autocorrelation radius: in thedirection perpendicular to a survey and in the direction of a survey (the latter

FIGURE 21.8 First row: Original patchy isotropic fields Second and third rows: Paths of

simulated surveys by parallel or zigzag transects and the reconstructed fields Right column

represents the gaps (the isolines are labeled by negative numbers).

represents, in fact, a part of the component of the autocorrelation radius) The sum

of these partial components, appearing as a result of all the transects, is equal to thefull component of the autocorrelation radius.21 Regarding the parallel pattern, theonly difference consists of the fact that the projection of the autocorrelation radiuscoincides with the radius itself Therefore, for both patterns, the average autocorre-lation radius can be calculated from the following formula:

FIGURE 21.9 First row: Original patchy anisotropic fields Second and third rows: Paths of

simulated surveys by parallel or zigzag transects conducted in the directions of major axes

of autocorrelation ellipses and the reconstructed fields Right column represents the gaps (the

isolines are labeled by negative numbers).

Rav=n + 1(Rp)nR,

where n is the number of transects, R and Rp are the autocorrelation radii in thedirection of a survey and in the perpendicular direction, respectively.

The results of simulated surveys with various ratios of the overall survey path

to the autocorrelation radius for the field (S/Rav

The existence of a distinct relationship fitting the generalized dataset (r2 vs S/Rav)confirms the possibility of using this ratio in determining the efficiency of acousticsurveys The results of the surveys conducted with the same overall paths but variousstarting points are presented in Figure 21.11 as columns As seen in the figure, the

FIGURE 21.10 First row: Original patchy rotated anisotropic fields Second and third rows:

Paths of simulated surveys by parallel or zigzag transects conducted in arbitrary directions

and the reconstructed fields Right column represents the gaps (the isolines are labeled by

negative numbers).

) are considered below (Figure 21.11)

regression curve regarding parallel transects is located everywhere below that forzigzag transects Therefore, the results obtained permit us to conclude that the formerallow, as a rule, less adequate reconstruction of a field than the latter do The result

of a specific mathematical experiment regarding parallel transects may be above thatfor zigzag transects This indicates that there is a great probability that the conclusion

is correct

Choice of the unit of sampling distance Let us consider the choice of this unit

in application to the parallel survey pattern The examples of original isotropic fields,

row, left, and middle); for the right field, the autocorrelation radius R = 0.16 Thefollowing distances between transects are taken as examples: D = 0.22 (left), D =0.14 (middle), or D = 0.24 (right) In the case of the left field, the increase of theunit of sampling distance from d = 1/50 to d = 1/7 results in lowering the coefficient

of determination from r2 = 0.83 (NSDP = 300, second row) to r2 = 0.74 (NSDP =

40, third row) For the middle field, the increase of the unit of sampling distancefrom d = 1/50 to d = 1/12 results in lowering the coefficient of determination from

r2 = 0.77 (NSDP = 450, second row) to r2 = 0.74 (NSDP = 104, third row) For the

FIGURE 21.11 Coefficient of determination between the reconstructed field and that

origi-nally generated as a function of the S/Rav ratio with regard to various survey patterns (left: parallel; right: zigzag) Filled symbols correspond to surveys of patchy distribution fields;

empty symbols: gappy fields Colors correspond to the results of simulated surveys of the

Figure 21.10, left; green: Figure 21.10, middle and right (surveys corresponding to the empty

dark blue, crimson, and yellow circles are not shown in Figure 21.8 to Figure 21.10) The coefficient of determination regarding isotropic fields is obtained by averaging the results

of the surveys carried out in the directions of abscissa and ordinate axes The solid line represents the nonlinear regression ensuring the minimal sum of squared residuals The dashed lines are the borders of the confidence interval with the probability of p = 0.99 The thick dashed line corresponds to the regression for the survey pattern shown in the opposite position

fields shown in Figure 21.8 to Figure 21.10 Dark blue: Figure 21.8, left; red: Figure 21.8, middle and right; crimson: Figure 21.9, left; light blue: Figure 21.9, middle and right; yellow:

given in Figure 21.1 (first row, left, and middle) are reproduced in Figure 21.12 (first

of the figure (left: zigzag; right: parallel) (See color insert following page 490.)